Introduction
Being able to measure medication adherence is hard. There are so many behavioral, social, and economic factors at play. Given the potential of using data to tell a patient’s medication story, where do we begin?
This post will explain our process in modeling medication prescriptions through a clinical measure working group (CLIME) and do so with an example of Prednisone.
The situation
Nearly $100-300 billion is lost annually when people don’t adhere to medication schedules, representing 3-10% of overall healthcare costs. This same report also showed that in recent years, the number of people taking three or more prescription drugs has doubled.
Medication adherence is a major concern across the healthcare industry: doing a deep dive into a clinical area alone is no fun and we wanted to increase the chances of us getting it right.
It takes a village to design a medication prescription schema
The strength of Open mHealth’s designed schemas is that they rely on clinical experts from different health domains as well as engineers working for products and services who share their health-specific data with the world. In order to crack the medication prescription nut, we invited multiple voices from around the industry because one of our guiding philosophies is that digital health data should be able to travel widely and be applied for many different uses while maintaining interoperability.
In the toolmaker corner we had Thomas Goetz from Iodine, Andy Bowline from NVOLVE, Alex Trahey from Wellframe, and Elliot Cohen from Pillpack.
Some of these toolmakers make consumer-facing applications, like Iodine’s drug reference library or Pillpack’s pharmacy services and medication adherence app. Others, like Wellframe, create industry-focused solutions that maximize the effectiveness of patient care, of which medication prescriptions are a major component.
On the clinical side, we had Kirby Lee from UCSF, Corrine Voils from Duke and Ian Kronish from Columbia University participate. Each of these clinical experts brought with them a patient-centric understanding on how we might design schemas to represent the complexities around tracking medication prescriptions.
As a side note, the med prescription CLIME was not responsible for figuring out a way to solve medication adherence. Instead, we worked together to model this complex data via schemas.
A model for understanding complex data
At first glance, medication prescription and adherence doesn’t seem so complex, right?
A doctor prescribes a patient drug X to be taken at frequency Y for Z amount of time. A pharmacist then fills the prescription. The patient picks it up, and then takes the medication at the assigned schedule. But the reality is far more complex.
Physicians, pharmacists, and patients are all active agents in understanding medication prescriptions: different schemas are needed to model the different roles.
How did we work through these multiple perspectives?
The easiest way was by providing concrete examples. In the case of medication prescriptions, we examined various medications, particularly ones that had complex dosing requirements. Our guiding principle at Open mHealth is always to make simple things easy, and complex things possible.
For this post, we’ll use the corticosteroid Prednisone.
It helps with pain, but not the pain of data integration
Before the CLIME process, we had already crafted multiple schemas, including a medication schema and a medication prescription schema, which model, respectively, a therapeutic agent (including name, strength and RxNORM code) and the dosing schedule for it.
For basic prescriptions (i.e., 1 tablet every day), the model is relatively easy. But what happens when the dosing is, for example, 1 tablet and ½ tablet taken on alternating days, or 1 tablet to be taken on Monday, Wednesday, Friday, and ½ tablet on Saturday? The medication prescription schema accommodates all these cases. [See: the sample data files entitled “valid medication prescription with sequence” and “valid medication prescription days of week” in the Sample Data section of the medication prescription schema]
From the CLIME discussions came the pharmacy medication dispensing schema. This models the pharmacist role, and includes the NDC code of the medication dispensed, the amount, the type of packaging used (i.e. a bottle vs. a dose pack), and the number of remaining refills.
By adding a pharmacy dispensing schema, we were one step closer to modeling the full prescription lifecycle and ensuring that stakeholders had the appropriate information. What needed to happen next? We modeled the medication prescription schema from the patient’s perspective.
The result was the patient medication schedule schema. It models how a patient integrates a prescription into their life schedule and includes the acceptable timeframe in which the medication should be taken, which is established by the physician and is based on indication and pharmacokinetics. For example, if a patient is supposed to take 1 tablet daily of Prednisone and s/he schedules it to take it at 8 AM, the acceptable window can be +/- 12 hours. In cases where a medicine must be taken at the same time every day (without an acceptable window), this data element can be left unspecified.
The taking of a dose is modeled in the single medication dose taken schema. When a patient doesn’t adhere to a given medication, the reason single medication dose not taken schema provides a way to record the reported reason for missing a dose. If a dose was taken but not at the required amounts, the single medication dose taken schema allows for the latter discrepancy to be recorded.
Far Reaching Medication Prescription Schemas
After we finished the CLIME, we reached out to the participants to see what their experience was like collaborating with everyone involved. The feedback was overwhelmingly positive.
Thomas Goetz of Iodine stated:
“The CLIME was an important step in bridging the gaps between technology companies, products and how clinicians and patients want to see data. More of these need to happen across lots of different measures.”
Similarly, Kirby Lee commented that,
“The Med Prescription CLIME got software developers, clinicians and researchers working together on better modeling of medication data. Our collaborative efforts have resulted in schemas that will facilitate the design of more effective and efficient solutions.”
Conclusion
The CLIME process was successful in that it demonstrated the effectiveness of an approach that brings together multiple parties with overlapping, but nevertheless distinct, interests. If we want to advance digital healthcare, creating more effective solutions, this kind of dialogue-driven open source data modeling is vital. With so many parties involved in a single medication prescription, missing even one rung on the ladder will make these solutions less effective.
At the same time, more money is flowing into the arena of medication prescriptions and adherence, whether from the federal government, grant-giving organizations, or VC funding. As a result, more companies want to create apps related to medication prescriptions and adherence.
Imagine, for instance, an app that tracks medication prescriptions integrates into an Electronic Health Record (EHR): that data needs to be translated into forms that make sense for clinicians and patients. Imagine a developer who wants a way for a patient to integrate prescription data from their Walgreens API and medication adherence data from another application.
Each data stream likely has its own language, so schemas would have to be created to translate this data into something that’s usable and useful to understand medication prescriptions. Our Open mHealth schemas, through careful modeling with many perspectives in this healthcare ecosystem, save the time, cost, and energy it would take to create schemas in-house.
Our schemas are never a means to an end, which is a net positive for any company interested in medication prescriptions. This is because the schemas allow any individual or entity to instead focus their energy on how to alleviate the multi-billion dollar burden that medication non-adherence puts on the healthcare industry.
On that note, browse the complete versions of our 10 medication prescription schemas and the set of sample data created to illustrate various cases. Let us know which ones you find most useful and why.
Give us some compelling new use cases we haven’t seen before. And more importantly, ask questions! We want you involved in this ongoing dialogue over medication prescriptions and medication adherence.